Spatial-Temporal Seizure Detection with Graph Attention Network and Bi-Directional Lstm Architecture

Abstract

The automatic detection of epileptic seizures by Electroencephalogram (EEG) can accelerate the diagnosis of the disease by neurologists, which is of incredible importance for the treatment of patients with epilepsy. However, current works on EEG-based seizure detection do not fully exploit the spatial-temporal information of EEG channels. In order to tackle this problem, we propose an automatic spatial-temporal epileptic seizure detection framework based on deep learning. Graph attention networks (GAT) are used as the front-end to extract spatial features. Thus, the topology of different EEG channels is fully exploited. Meanwhile, bi-directional long short-term memory (BiLSTM) network is used as the back-end to mine time relations and making the final decision according to the state before and after the current moment. Experiments are conducted on the CHB-MIT dataset. The seizure detection accuracy, sensitivity, and specificity on the CHB-MIT dataset are 98.52\%, 97.75\%, and 94.34\%, respectively. Extensive experiments on CHB-MIT dataset demonstrate that the model can effectively detect seizures from the raw EEG signals without extra feature extraction. The performance of the model on the CHB-MIT database is better than or comparable to the-state-of-the-arts.

Keywords: Seizure detection, Scalp EEG, Deep learning, graph attention network, Bi-directional LSTM